1. Field of the Invention
The present invention relates to a character recognition technology, and more particularly, a system and method for recognizing information of a vehicle license plate.
2. Discussion of Related Art
A vehicle license plate recognition technology is a technology that automatically recognizes character string information of a license plate attached to a vehicle from an image, that is, a core technology widely used in speed control, parking lot administration, road toll collection, and security closed-circuit television (CCTV).
Generally, acquiring character information of a vehicle license plate requires three steps of processing, including license plate detection, character separation, and character recognition. In the license plate detection, a position of a license plate is roughly estimated, in the character separation, positions of individual characters are accurately estimated, and in the character recognition, the individual characters are recognized on the basis of the estimated positions.
According to a method of recognizing a vehicle license plate, a position of a license plate is roughly detected from an image of a vehicle, and from the detected position of the license plate, a position of each character of the license plate and a license plate type are determined. Then, a license plate character image is extracted from position information of each character so that a character string of the license plate is reconstructed.
In the case of license plate detection, the position of the license plate is detected by analyzing an image, but a region of interest (ROI) may be provided in advance as in a parking lot entrance, or a position of a vehicle may be provided by an external sensor (e.g., a loop sensor) as in speed control so that the license plate detection can be achieved.
In the case of character separation, a position of each character is estimated by analyzing a whole image, and position information of each character is obtained by extracting edge information of each character or a license plate using histogram smoothing, vertical/horizontal projection, template, binarization, and the like. Among vehicle license plate recognition related technologies, the character separation is the most actively developed technology, and the license plate type is determined mainly in the character separation.
In the case of character information recognition, a license plate character string is constructed by processing license plate type information and individual character images, which are determined in the character separation, through a type-specific recognizer. In this case, a recognizer corresponding to a license plate type among type-specific recognizers is selected and applied so that individual characters are recognized. Character information recognition is mainly achieved by a method of comparing a character image stored in a template scheme or a method of training a recognizer with learning data through a neural network.
In this regard, a technology is disclosed in Korean Patent Registration Publication No. 10-0802605 (title: Apparatus and Method of Recognizing Numerals from Vehicle License Plate) including a camera configured to capture an image of a license plate of a vehicle; a binary processing unit configured to convert a gray level of a pixel of the image of the license plate of the vehicle that is output from the camera into a binary value such that a gray level that is higher than or equal to a predetermined value is converted into a binary value corresponding to white and a gray level that is lower than the predetermined value is converted into a binary value corresponding to black; a micro line processing unit configured to extract a micro line corresponding to a region of pieces of adjacent pixel data binarized by the binarization processor which have the same value of black or white; a scale calculating unit configured to calculate a value of a micro line belonging to a main parameter and a sub parameter based on main parameters determined as a vertical line, a horizontal line, a slanted vertical line, a curve, and a circle and sub parameters subdivided as a horizontal line on the slanted vertical line, a horizontal line under the slanted vertical line, a curve on the slanted vertical line, a curve under the slanted vertical line, a circle on the slanted vertical line, or a circle under the slanted vertical line with respect to the slanted vertical line in the main parameters to extract a feature element that distinguishes numbers 0 to 9 from each other; a number recognizing unit configured to determine a number in the image of the vehicle license plate from the value belonging to the main parameter and the sub parameter calculated by the standard calculator.
However, in a conventional license plate recognition method, characteristics of characters according to a specific condition of a vehicle license plate are rarely used, and even when the license plate type information processed in the character separation is incorrect, a recognition result is output, thus leading to a misreading. When the misread result is used as it is, processing costs for correcting the misread result are higher than a cost for a rejection processing that does not output a result.
In order to reduce misreading, a method of increasing reliability of a recognizer may be used, but this method has a trade-off of lower throughput for higher reliability. That is, when reliability is increased, the recognition rate is lowered due to an increase in the number of representatives to be processed. Meanwhile, when the character separation process outputs erroneous type information of a license plate at a higher accuracy, a result of recognition is accordingly misprocessed, and incorrect license plate information is output (misread). Such an output of incorrect license plate information may result in higher costs than the cost required for a rejection that does not output a recognition result.